13D.2 Identifying the ITCZ in satellite data using Markov Random Fields: Overview and interannual variability in the east Pacific

Thursday, 13 May 2010: 10:30 AM
Tucson Salon A-C (JW MArriott Starr Pass Resort)
Gudrun Magnusdottir, University of California, Irvine, Irvine, CA; and C. Bain, J. De Paz Rodrigues, J. Kramer, P. Smyth, and H. Stern

A team of interdisciplinary scientists at the University of California Irvine has developed a statistical model (hereafter LIMA) for labeling the Intertropical Convergence Zone (ITCZ) in satellite data. LIMA relies on a probabilistic framework that incorporates observed satellite data, a latent Markov random field that represents the presence/absence of the ITCZ, and “a priori” information about the most likely spatial location of ITCZ formation. The focus here is on the tropical E Pacific (90W-180W, equator to 20N) during the summer half year (May-October). In this presentation we focus on methodology and interannual variability of the ITCZ. A companion presentation (Bain et al) will address intraseasonal variability and extratropical connections.

The statistical model allows us to emulate the human as she views several instantaneous data fields over several consecutive days of conditions over the region of interest and decides which (if any) regions are representative of the ITCZ. The process is made entirely automatic after learning the parameters of the statistical model, which requires knowledge of what values of the satellite fields are characteristic of ITCZ and non-ITCZ events. This information was obtained by manually labeling ITCZ regions (if any are present) based on three different sets of satellite data for one month. The prior of most likely ITCZ location is a zonally symmetric strip centered at 10N. The model can easily and efficiently deal with missing data as well as data of different spatial and temporal resolutions.

The statistical model is novel as is the long record of GOES measurements that NOAA has only recently made available in a format convenient for climate research. The following satellite data were used:

1-2) GOES visible (VIS) and infrared (IR) images made available in the E Pacific basin by Ken Knapp (NOAA's National Climate Center). They were originally derived from ISCCP B1 data and are of 0.07 degree spatial resolution that we coarsen to 0.5 degree resolution to increase model efficiency.

1) The IR data are available every three hours from 1980-2008 except for 1988 and 1989.

2) The VIS data are available for the entire region once per day, at 2100 UTC. The early part of the record turned out to be too noisy for our application. We use 1995-2008.

3) Total precipitable water (TPW) images made available by Deborah Smith (Remote Sensing Systems Inc.). They are a composite of all available microwave data (e.g., SSM/I, TMI, AMSR-E) of total column water vapor. The data are of 1/4 degree spatial resolution that we coarsen to 0.5 degrees and are available twice per day at 0600 and 1800 UTC from 1988.

The latitudinal tilt of the ITCZ is evident in our dataset (northward tilt, from west to east), which is consistent with idealized modeling studies as well as results from ERA-40 reanalysis (using low level vorticity). The E Pacific ITCZ may be separated by longitude into three equal length sections. The easternmost section (90-120W) is most affected by tropical cyclones and a part of this section was the subject of a field program (EPIC) in September 2001. Consistent with results from EPIC this section of the ITCZ tends to be located on the southern side of the local oceanic warm pool. However, the westernmost section of the ITCZ (150-180W) is thinner in latitude and located on the northern side of the local oceanic warm pool. Strong El Nino years shift the ITCZ drastically in latitude, however the total area of the ITCZ and the depth of convection is not greatly affected. We shall discuss 1) interannual variability of the E Pacific ITCZ in this long time series, 2) possible applications of the statistical model to other areas of the tropical oceans, 3) the use of other datasets (such as output from Global Climate models (GCMs)) as inputs to the statistical model where LIMA could potentially be an important diagnostic tool that could prove valuable for GCM development.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner